subagent-driven-development
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill implements a workflow that ingests task descriptions from plan files and interpolates them into subagent prompts. This creates a surface where instructions embedded in a plan could potentially influence subagent execution.\n
- Ingestion points: Task text from plan files is pasted into
implementer-prompt.mdandspec-reviewer-prompt.md.\n - Boundary markers: The prompts utilize markdown headers (e.g.,
## Task Description) to separate task content from instructions.\n - Capability inventory: Subagents are tasked with file modifications, running tests, and performing git commits.\n
- Sanitization: The workflow relies on the multi-stage review process (spec reviewer and code quality reviewer) rather than programmatic sanitization to ensure implementation matches intentions.\n- [COMMAND_EXECUTION]: The workflow requires the implementation subagent to execute tests and perform git commits, implying the capability to run shell commands in the development environment. This is an intended feature of the development-driven skill.
Audit Metadata